PROJECT SUMMARY Molecular profiles of tumors are nowadays used to determine prognosis and to guide therapy. For example the presence of a mutation in the EGFR gene will most likely lead to anti-EGFR therapy. Recently an image phenotype was discovered that acts as a biomarker of EGFR mutation. This is a precursor of the possibilities of a new emerging field called radiogenomics defined as directly linking imaging features to underlying molecular properties. Research in radiogenomics is rapidly gaining recognition as a powerful new field that has several promising applications, such as non-invasive molecular lesion assessment. When image surrogates can be identified that mirror relevant molecular aberrations (e.g. a mutation in the EGFR gene) they can be readily translated in clinical care. The value added by radiogenomics can be readily translated, as medical imaging is part of routine management in oncology. However, these early applications have not taken full advantage of the opportunities. First, they limit the correlation to either a handful of manually annotated image features and a pre-selected set of molecular parameters. Secondly, the initial applications are limited to a single omics by focusing on gene expression, without taking into account DNA mutations, DNA copy number changes or DNA methylation changes. We will develop a radiogenomics framework to identify non-invasive biomarkers mirroring relevant molecular tumor properties that impact treatment and clinical outcome of human brain tumors. Our objective is not to mimic a radiologist's expertise through computational means, but to empower radiologists and clinicians with new biomarkers. We will offer innovative new algorithms to represent medical images. Once such a representation is computed (e.g., in the form of a large data matrix), we will identify univariate and multivariate image signatures predictive of clinical outcome. Next, we will use sophisticated methods for integration with molecular data to interrogate different views of the data with respect to a clinically relevant outcome. The end result is a radiogenomics map where image signatures of molecular properties and tumor heterogeneity can be hypothesized and validated. We will have image signatures that are prognostic and image signatures reflecting actionable molecular properties of a tumor such as drug target activity or drug signatures.